lustre/ptlrpc/lproc_ptlrpc.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2016, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  */
  32 #define DEBUG_SUBSYSTEM S_CLASS
  33
  34
  35 #include <obd_support.h>
  36 #include <obd.h>
  37 #include <lprocfs_status.h>
  38 #include <lustre/lustre_idl.h>
  39 #include <lustre_net.h>
  40 #include <obd_class.h>
  41 #include "ptlrpc_internal.h"
  42
  43
  44 static struct ll_rpc_opcode {
  45      __u32       opcode;
  46      const char *opname;
  47 } ll_rpc_opcode_table[LUSTRE_MAX_OPCODES] = {
  48         { OST_REPLY,        "ost_reply" },
  49         { OST_GETATTR,      "ost_getattr" },
  50         { OST_SETATTR,      "ost_setattr" },
  51         { OST_READ,         "ost_read" },
  52         { OST_WRITE,        "ost_write" },
  53         { OST_CREATE ,      "ost_create" },
  54         { OST_DESTROY,      "ost_destroy" },
  55         { OST_GET_INFO,     "ost_get_info" },
  56         { OST_CONNECT,      "ost_connect" },
  57         { OST_DISCONNECT,   "ost_disconnect" },
  58         { OST_PUNCH,        "ost_punch" },
  59         { OST_OPEN,         "ost_open" },
  60         { OST_CLOSE,        "ost_close" },
  61         { OST_STATFS,       "ost_statfs" },
  62         { 14,                NULL },    /* formerly OST_SAN_READ */
  63         { 15,                NULL },    /* formerly OST_SAN_WRITE */
  64         { OST_SYNC,         "ost_sync" },
  65         { OST_SET_INFO,     "ost_set_info" },
  66         { OST_QUOTACHECK,   "ost_quotacheck" },
  67         { OST_QUOTACTL,     "ost_quotactl" },
  68         { OST_QUOTA_ADJUST_QUNIT, "ost_quota_adjust_qunit" },
  69         { OST_LADVISE,      "ost_ladvise" },
  70         { MDS_GETATTR,      "mds_getattr" },
  71         { MDS_GETATTR_NAME, "mds_getattr_lock" },
  72         { MDS_CLOSE,        "mds_close" },
  73         { MDS_REINT,        "mds_reint" },
  74         { MDS_READPAGE,     "mds_readpage" },
  75         { MDS_CONNECT,      "mds_connect" },
  76         { MDS_DISCONNECT,   "mds_disconnect" },
  77         { MDS_GET_ROOT,     "mds_get_root" },
  78         { MDS_STATFS,       "mds_statfs" },
  79         { MDS_PIN,          "mds_pin" },
  80         { MDS_UNPIN,        "mds_unpin" },
  81         { MDS_SYNC,         "mds_sync" },
  82         { MDS_DONE_WRITING, "mds_done_writing" },
  83         { MDS_SET_INFO,     "mds_set_info" },
  84         { MDS_QUOTACHECK,   "mds_quotacheck" },
  85         { MDS_QUOTACTL,     "mds_quotactl" },
  86         { MDS_GETXATTR,     "mds_getxattr" },
  87         { MDS_SETXATTR,     "mds_setxattr" },
  88         { MDS_WRITEPAGE,    "mds_writepage" },
  89         { MDS_IS_SUBDIR,    "mds_is_subdir" },
  90         { MDS_GET_INFO,     "mds_get_info" },
  91         { MDS_HSM_STATE_GET, "mds_hsm_state_get" },
  92         { MDS_HSM_STATE_SET, "mds_hsm_state_set" },
  93         { MDS_HSM_ACTION,   "mds_hsm_action" },
  94         { MDS_HSM_PROGRESS, "mds_hsm_progress" },
  95         { MDS_HSM_REQUEST,  "mds_hsm_request" },
  96         { MDS_HSM_CT_REGISTER, "mds_hsm_ct_register" },
  97         { MDS_HSM_CT_UNREGISTER, "mds_hsm_ct_unregister" },
  98         { MDS_SWAP_LAYOUTS,     "mds_swap_layouts" },
  99         { LDLM_ENQUEUE,     "ldlm_enqueue" },
 100         { LDLM_CONVERT,     "ldlm_convert" },
 101         { LDLM_CANCEL,      "ldlm_cancel" },
 102         { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
 103         { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
 104         { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
 105         { LDLM_SET_INFO,    "ldlm_set_info" },
 106         { MGS_CONNECT,      "mgs_connect" },
 107         { MGS_DISCONNECT,   "mgs_disconnect" },
 108         { MGS_EXCEPTION,    "mgs_exception" },
 109         { MGS_TARGET_REG,   "mgs_target_reg" },
 110         { MGS_TARGET_DEL,   "mgs_target_del" },
 111         { MGS_SET_INFO,     "mgs_set_info" },
 112         { MGS_CONFIG_READ,  "mgs_config_read" },
 113         { OBD_PING,         "obd_ping" },
 114         { OBD_LOG_CANCEL,       "llog_cancel" },
 115         { OBD_QC_CALLBACK,  "obd_quota_callback" },
 116         { OBD_IDX_READ,     "dt_index_read" },
 117         { LLOG_ORIGIN_HANDLE_CREATE,     "llog_origin_handle_open" },
 118         { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
 119         { LLOG_ORIGIN_HANDLE_READ_HEADER,"llog_origin_handle_read_header" },
 120         { LLOG_ORIGIN_HANDLE_WRITE_REC,  "llog_origin_handle_write_rec" },
 121         { LLOG_ORIGIN_HANDLE_CLOSE,      "llog_origin_handle_close" },
 122         { LLOG_ORIGIN_CONNECT,           "llog_origin_connect" },
 123         { LLOG_CATINFO,                  "llog_catinfo" },
 124         { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
 125         { LLOG_ORIGIN_HANDLE_DESTROY,    "llog_origin_handle_destroy" },
 126         { QUOTA_DQACQ,      "quota_acquire" },
 127         { QUOTA_DQREL,      "quota_release" },
 128         { SEQ_QUERY,        "seq_query" },
 129         { SEC_CTX_INIT,     "sec_ctx_init" },
 130         { SEC_CTX_INIT_CONT,"sec_ctx_init_cont" },
 131         { SEC_CTX_FINI,     "sec_ctx_fini" },
 132         { FLD_QUERY,        "fld_query" },
 133         { FLD_READ,         "fld_read" },
 134         { OUT_UPDATE,       "out_update" },
 135         { LFSCK_NOTIFY,     "lfsck_notify" },
 136         { LFSCK_QUERY,      "lfsck_query" },
 137 };
 138
 139 static struct ll_eopcode {
 140      __u32       opcode;
 141      const char *opname;
 142 } ll_eopcode_table[EXTRA_LAST_OPC] = {
 143         { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" },
 144         { LDLM_PLAIN_ENQUEUE,   "ldlm_plain_enqueue" },
 145         { LDLM_EXTENT_ENQUEUE,  "ldlm_extent_enqueue" },
 146         { LDLM_FLOCK_ENQUEUE,   "ldlm_flock_enqueue" },
 147         { LDLM_IBITS_ENQUEUE,   "ldlm_ibits_enqueue" },
 148         { MDS_REINT_SETATTR,    "mds_reint_setattr" },
 149         { MDS_REINT_CREATE,     "mds_reint_create" },
 150         { MDS_REINT_LINK,       "mds_reint_link" },
 151         { MDS_REINT_UNLINK,     "mds_reint_unlink" },
 152         { MDS_REINT_RENAME,     "mds_reint_rename" },
 153         { MDS_REINT_OPEN,       "mds_reint_open" },
 154         { MDS_REINT_SETXATTR,   "mds_reint_setxattr" },
 155         { BRW_READ_BYTES,       "read_bytes" },
 156         { BRW_WRITE_BYTES,      "write_bytes" },
 157 };
 158
 159 const char *ll_opcode2str(__u32 opcode)
 160 {
 161         /* When one of the assertions below fail, chances are that:
 162          *     1) A new opcode was added in include/lustre/lustre_idl.h,
 163          *        but is missing from the table above.
 164          * or  2) The opcode space was renumbered or rearranged,
 165          *        and the opcode_offset() function in
 166          *        ptlrpc_internal.h needs to be modified.
 167          */
 168         __u32 offset = opcode_offset(opcode);
 169         LASSERTF(offset < LUSTRE_MAX_OPCODES,
 170                  "offset %u >= LUSTRE_MAX_OPCODES %u\n",
 171                  offset, LUSTRE_MAX_OPCODES);
 172         LASSERTF(ll_rpc_opcode_table[offset].opcode == opcode,
 173                  "ll_rpc_opcode_table[%u].opcode %u != opcode %u\n",
 174                  offset, ll_rpc_opcode_table[offset].opcode, opcode);
 175         return ll_rpc_opcode_table[offset].opname;
 176 }
 177
 178 static const char *ll_eopcode2str(__u32 opcode)
 179 {
 180         LASSERT(ll_eopcode_table[opcode].opcode == opcode);
 181         return ll_eopcode_table[opcode].opname;
 182 }
 183
 184 #ifdef CONFIG_PROC_FS
 185 static void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
 186                              char *name, struct proc_dir_entry **procroot_ret,
 187                              struct lprocfs_stats **stats_ret)
 188 {
 189         struct proc_dir_entry *svc_procroot;
 190         struct lprocfs_stats *svc_stats;
 191         int i, rc;
 192         unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
 193                                           LPROCFS_CNTR_STDDEV;
 194
 195         LASSERT(*procroot_ret == NULL);
 196         LASSERT(*stats_ret == NULL);
 197
 198         svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES+LUSTRE_MAX_OPCODES,0);
 199         if (svc_stats == NULL)
 200                 return;
 201
 202         if (dir) {
 203                 svc_procroot = lprocfs_register(dir, root, NULL, NULL);
 204                 if (IS_ERR(svc_procroot)) {
 205                         lprocfs_free_stats(&svc_stats);
 206                         return;
 207                 }
 208         } else {
 209                 svc_procroot = root;
 210         }
 211
 212         lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
 213                              svc_counter_config, "req_waittime", "usec");
 214         lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
 215                              svc_counter_config, "req_qdepth", "reqs");
 216         lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
 217                              svc_counter_config, "req_active", "reqs");
 218         lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
 219                              svc_counter_config, "req_timeout", "sec");
 220         lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
 221                              svc_counter_config, "reqbuf_avail", "bufs");
 222         for (i = 0; i < EXTRA_LAST_OPC; i++) {
 223                 char *units;
 224
 225                 switch(i) {
 226                 case BRW_WRITE_BYTES:
 227                 case BRW_READ_BYTES:
 228                         units = "bytes";
 229                         break;
 230                 default:
 231                         units = "reqs";
 232                         break;
 233                 }
 234                 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
 235                                      svc_counter_config,
 236                                      ll_eopcode2str(i), units);
 237         }
 238         for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
 239                 __u32 opcode = ll_rpc_opcode_table[i].opcode;
 240                 lprocfs_counter_init(svc_stats,
 241                                      EXTRA_MAX_OPCODES + i, svc_counter_config,
 242                                      ll_opcode2str(opcode), "usec");
 243         }
 244
 245         rc = lprocfs_register_stats(svc_procroot, name, svc_stats);
 246         if (rc < 0) {
 247                 if (dir)
 248                         lprocfs_remove(&svc_procroot);
 249                 lprocfs_free_stats(&svc_stats);
 250         } else {
 251                 if (dir)
 252                         *procroot_ret = svc_procroot;
 253                 *stats_ret = svc_stats;
 254         }
 255 }
 256
 257 static int
 258 ptlrpc_lprocfs_req_history_len_seq_show(struct seq_file *m, void *v)
 259 {
 260         struct ptlrpc_service *svc = m->private;
 261         struct ptlrpc_service_part *svcpt;
 262         int     total = 0;
 263         int     i;
 264
 265         ptlrpc_service_for_each_part(svcpt, i, svc)
 266                 total += svcpt->scp_hist_nrqbds;
 267
 268         seq_printf(m, "%d\n", total);
 269         return 0;
 270 }
 271 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_req_history_len);
 272
 273 static int
 274 ptlrpc_lprocfs_req_history_max_seq_show(struct seq_file *m, void *n)
 275 {
 276         struct ptlrpc_service *svc = m->private;
 277         struct ptlrpc_service_part *svcpt;
 278         int     total = 0;
 279         int     i;
 280
 281         ptlrpc_service_for_each_part(svcpt, i, svc)
 282                 total += svc->srv_hist_nrqbds_cpt_max;
 283
 284         seq_printf(m, "%d\n", total);
 285         return 0;
 286 }
 287
 288 static ssize_t
 289 ptlrpc_lprocfs_req_history_max_seq_write(struct file *file,
 290                                          const char __user *buffer,
 291                                          size_t count, loff_t *off)
 292 {
 293         struct seq_file *m = file->private_data;
 294         struct ptlrpc_service *svc = m->private;
 295         int bufpages;
 296         __s64 val;
 297         int rc;
 298
 299         rc = lprocfs_str_to_s64(buffer, count, &val);
 300         if (rc < 0)
 301                 return rc;
 302
 303         if (val < 0 || val > INT_MAX)
 304                 return -ERANGE;
 305
 306         /* This sanity check is more of an insanity check; we can still
 307          * hose a kernel by allowing the request history to grow too
 308          * far. */
 309         bufpages = (svc->srv_buf_size + PAGE_SIZE - 1) >>
 310                                                         PAGE_SHIFT;
 311         if (val > totalram_pages/(2 * bufpages))
 312                 return -ERANGE;
 313
 314         spin_lock(&svc->srv_lock);
 315
 316         if (val == 0)
 317                 svc->srv_hist_nrqbds_cpt_max = 0;
 318         else
 319                 svc->srv_hist_nrqbds_cpt_max =
 320                         max(1, ((int)val / svc->srv_ncpts));
 321
 322         spin_unlock(&svc->srv_lock);
 323
 324         return count;
 325 }
 326 LPROC_SEQ_FOPS(ptlrpc_lprocfs_req_history_max);
 327
 328 static int
 329 ptlrpc_lprocfs_threads_min_seq_show(struct seq_file *m, void *n)
 330 {
 331         struct ptlrpc_service *svc = m->private;
 332
 333         seq_printf(m, "%d\n",
 334                    svc->srv_nthrs_cpt_init * svc->srv_ncpts);
 335         return 0;
 336 }
 337
 338 static ssize_t
 339 ptlrpc_lprocfs_threads_min_seq_write(struct file *file,
 340                                      const char __user *buffer,
 341                                      size_t count, loff_t *off)
 342 {
 343         struct seq_file *m = file->private_data;
 344         struct ptlrpc_service *svc = m->private;
 345         __s64 val;
 346         int rc = lprocfs_str_to_s64(buffer, count, &val);
 347
 348         if (rc < 0)
 349                 return rc;
 350
 351         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 352                 return -ERANGE;
 353
 354         spin_lock(&svc->srv_lock);
 355         if (val > svc->srv_nthrs_cpt_limit * svc->srv_ncpts) {
 356                 spin_unlock(&svc->srv_lock);
 357                 return -ERANGE;
 358         }
 359
 360         svc->srv_nthrs_cpt_init = (int)val / svc->srv_ncpts;
 361
 362         spin_unlock(&svc->srv_lock);
 363
 364         return count;
 365 }
 366 LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_min);
 367
 368 static int
 369 ptlrpc_lprocfs_threads_started_seq_show(struct seq_file *m, void *n)
 370 {
 371         struct ptlrpc_service           *svc = m->private;
 372         struct ptlrpc_service_part      *svcpt;
 373         int     total = 0;
 374         int     i;
 375
 376         ptlrpc_service_for_each_part(svcpt, i, svc)
 377                 total += svcpt->scp_nthrs_running;
 378
 379         seq_printf(m, "%d\n", total);
 380         return 0;
 381 }
 382 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_threads_started);
 383
 384 static int
 385 ptlrpc_lprocfs_threads_max_seq_show(struct seq_file *m, void *n)
 386 {
 387         struct ptlrpc_service *svc = m->private;
 388
 389         seq_printf(m, "%d\n",
 390                    svc->srv_nthrs_cpt_limit * svc->srv_ncpts);
 391         return 0;
 392 }
 393
 394 static ssize_t
 395 ptlrpc_lprocfs_threads_max_seq_write(struct file *file,
 396                                      const char __user *buffer,
 397                                      size_t count, loff_t *off)
 398 {
 399         struct seq_file *m = file->private_data;
 400         struct ptlrpc_service *svc = m->private;
 401         __s64 val;
 402         int rc = lprocfs_str_to_s64(buffer, count, &val);
 403
 404         if (rc < 0)
 405                 return rc;
 406
 407         if (val / svc->srv_ncpts < PTLRPC_NTHRS_INIT)
 408                 return -ERANGE;
 409
 410         spin_lock(&svc->srv_lock);
 411         if (val < svc->srv_nthrs_cpt_init * svc->srv_ncpts) {
 412                 spin_unlock(&svc->srv_lock);
 413                 return -ERANGE;
 414         }
 415
 416         svc->srv_nthrs_cpt_limit = (int)val / svc->srv_ncpts;
 417
 418         spin_unlock(&svc->srv_lock);
 419
 420         return count;
 421 }
 422 LPROC_SEQ_FOPS(ptlrpc_lprocfs_threads_max);
 423
 424 /**
 425  * Translates \e ptlrpc_nrs_pol_state values to human-readable strings.
 426  *
 427  * \param[in] state The policy state
 428  */
 429 static const char *nrs_state2str(enum ptlrpc_nrs_pol_state state)
 430 {
 431         switch (state) {
 432         default:
 433                 LBUG();
 434         case NRS_POL_STATE_INVALID:
 435                 return "invalid";
 436         case NRS_POL_STATE_STOPPED:
 437                 return "stopped";
 438         case NRS_POL_STATE_STOPPING:
 439                 return "stopping";
 440         case NRS_POL_STATE_STARTING:
 441                 return "starting";
 442         case NRS_POL_STATE_STARTED:
 443                 return "started";
 444         }
 445 }
 446
 447 /**
 448  * Obtains status information for \a policy.
 449  *
 450  * Information is copied in \a info.
 451  *
 452  * \param[in] policy The policy
 453  * \param[out] info  Holds returned status information
 454  */
 455 void nrs_policy_get_info_locked(struct ptlrpc_nrs_policy *policy,
 456                                 struct ptlrpc_nrs_pol_info *info)
 457 {
 458         LASSERT(policy != NULL);
 459         LASSERT(info != NULL);
 460         assert_spin_locked(&policy->pol_nrs->nrs_lock);
 461
 462         LASSERT(sizeof(info->pi_arg) == sizeof(policy->pol_arg));
 463         memcpy(info->pi_name, policy->pol_desc->pd_name, NRS_POL_NAME_MAX);
 464         memcpy(info->pi_arg, policy->pol_arg, sizeof(policy->pol_arg));
 465
 466         info->pi_fallback    = !!(policy->pol_flags & PTLRPC_NRS_FL_FALLBACK);
 467         info->pi_state       = policy->pol_state;
 468         /**
 469          * XXX: These are accessed without holding
 470          * ptlrpc_service_part::scp_req_lock.
 471          */
 472         info->pi_req_queued  = policy->pol_req_queued;
 473         info->pi_req_started = policy->pol_req_started;
 474 }
 475
 476 /**
 477  * Reads and prints policy status information for all policies of a PTLRPC
 478  * service.
 479  */
 480 static int ptlrpc_lprocfs_nrs_seq_show(struct seq_file *m, void *n)
 481 {
 482         struct ptlrpc_service          *svc = m->private;
 483         struct ptlrpc_service_part     *svcpt;
 484         struct ptlrpc_nrs              *nrs;
 485         struct ptlrpc_nrs_policy       *policy;
 486         struct ptlrpc_nrs_pol_info     *infos;
 487         struct ptlrpc_nrs_pol_info      tmp;
 488         unsigned                        num_pols;
 489         unsigned                        pol_idx = 0;
 490         bool                            hp = false;
 491         int                             i;
 492         int                             rc = 0;
 493         ENTRY;
 494
 495         /**
 496          * Serialize NRS core lprocfs operations with policy registration/
 497          * unregistration.
 498          */
 499         mutex_lock(&nrs_core.nrs_mutex);
 500
 501         /**
 502          * Use the first service partition's regular NRS head in order to obtain
 503          * the number of policies registered with NRS heads of this service. All
 504          * service partitions will have the same number of policies.
 505          */
 506         nrs = nrs_svcpt2nrs(svc->srv_parts[0], false);
 507
 508         spin_lock(&nrs->nrs_lock);
 509         num_pols = svc->srv_parts[0]->scp_nrs_reg.nrs_num_pols;
 510         spin_unlock(&nrs->nrs_lock);
 511
 512         OBD_ALLOC(infos, num_pols * sizeof(*infos));
 513         if (infos == NULL)
 514                 GOTO(out, rc = -ENOMEM);
 515 again:
 516
 517         ptlrpc_service_for_each_part(svcpt, i, svc) {
 518                 nrs = nrs_svcpt2nrs(svcpt, hp);
 519                 spin_lock(&nrs->nrs_lock);
 520
 521                 pol_idx = 0;
 522
 523                 list_for_each_entry(policy, &nrs->nrs_policy_list,
 524                                     pol_list) {
 525                         LASSERT(pol_idx < num_pols);
 526
 527                         nrs_policy_get_info_locked(policy, &tmp);
 528                         /**
 529                          * Copy values when handling the first service
 530                          * partition.
 531                          */
 532                         if (i == 0) {
 533                                 memcpy(infos[pol_idx].pi_name, tmp.pi_name,
 534                                        NRS_POL_NAME_MAX);
 535                                 memcpy(infos[pol_idx].pi_arg, tmp.pi_arg,
 536                                        sizeof(tmp.pi_arg));
 537                                 memcpy(&infos[pol_idx].pi_state, &tmp.pi_state,
 538                                        sizeof(tmp.pi_state));
 539                                 infos[pol_idx].pi_fallback = tmp.pi_fallback;
 540                                 /**
 541                                  * For the rest of the service partitions
 542                                  * sanity-check the values we get.
 543                                  */
 544                         } else {
 545                                 LASSERT(strncmp(infos[pol_idx].pi_name,
 546                                                 tmp.pi_name,
 547                                                 NRS_POL_NAME_MAX) == 0);
 548                                 LASSERT(strncmp(infos[pol_idx].pi_arg,
 549                                                 tmp.pi_arg,
 550                                                 sizeof(tmp.pi_arg)) == 0);
 551                                 /**
 552                                  * Not asserting ptlrpc_nrs_pol_info::pi_state,
 553                                  * because it may be different between
 554                                  * instances of the same policy in different
 555                                  * service partitions.
 556                                  */
 557                                 LASSERT(infos[pol_idx].pi_fallback ==
 558                                         tmp.pi_fallback);
 559                         }
 560
 561                         infos[pol_idx].pi_req_queued += tmp.pi_req_queued;
 562                         infos[pol_idx].pi_req_started += tmp.pi_req_started;
 563
 564                         pol_idx++;
 565                 }
 566                 spin_unlock(&nrs->nrs_lock);
 567         }
 568
 569         /**
 570          * Policy status information output is in YAML format.
 571          * For example:
 572          *
 573          *      regular_requests:
 574          *        - name: fifo
 575          *          state: started
 576          *          fallback: yes
 577          *          queued: 0
 578          *          active: 0
 579          *
 580          *        - name: crrn
 581          *          state: started
 582          *          fallback: no
 583          *          queued: 2015
 584          *          active: 384
 585          *
 586          *      high_priority_requests:
 587          *        - name: fifo
 588          *          state: started
 589          *          fallback: yes
 590          *          queued: 0
 591          *          active: 2
 592          *
 593          *        - name: crrn
 594          *          state: stopped
 595          *          fallback: no
 596          *          queued: 0
 597          *          active: 0
 598          */
 599         seq_printf(m, "%s\n", !hp ? "\nregular_requests:" :
 600                    "high_priority_requests:");
 601
 602         for (pol_idx = 0; pol_idx < num_pols; pol_idx++) {
 603                 if (strlen(infos[pol_idx].pi_arg) > 0)
 604                         seq_printf(m, "  - name: %s %s\n",
 605                                    infos[pol_idx].pi_name,
 606                                    infos[pol_idx].pi_arg);
 607                 else
 608                         seq_printf(m, "  - name: %s\n",
 609                                    infos[pol_idx].pi_name);
 610
 611
 612                 seq_printf(m, "    state: %s\n"
 613                            "    fallback: %s\n"
 614                            "    queued: %-20d\n"
 615                            "    active: %-20d\n\n",
 616                            nrs_state2str(infos[pol_idx].pi_state),
 617                            infos[pol_idx].pi_fallback ? "yes" : "no",
 618                            (int)infos[pol_idx].pi_req_queued,
 619                            (int)infos[pol_idx].pi_req_started);
 620         }
 621
 622         if (!hp && nrs_svc_has_hp(svc)) {
 623                 memset(infos, 0, num_pols * sizeof(*infos));
 624
 625                 /**
 626                  * Redo the processing for the service's HP NRS heads' policies.
 627                  */
 628                 hp = true;
 629                 goto again;
 630         }
 631
 632 out:
 633         if (infos)
 634                 OBD_FREE(infos, num_pols * sizeof(*infos));
 635
 636         mutex_unlock(&nrs_core.nrs_mutex);
 637
 638         RETURN(rc);
 639 }
 640
 641
 642 #define LPROCFS_NRS_WR_MAX_ARG (1024)
 643 /**
 644  * The longest valid command string is the maxium policy name size, plus the
 645  * length of the " reg" substring, plus the lenght of argument
 646  */
 647 #define LPROCFS_NRS_WR_MAX_CMD  (NRS_POL_NAME_MAX + sizeof(" reg") - 1 \
 648                                  + LPROCFS_NRS_WR_MAX_ARG)
 649
 650 /**
 651  * Starts and stops a given policy on a PTLRPC service.
 652  *
 653  * Commands consist of the policy name, followed by an optional [reg|hp] token;
 654  * if the optional token is omitted, the operation is performed on both the
 655  * regular and high-priority (if the service has one) NRS head.
 656  */
 657 static ssize_t
 658 ptlrpc_lprocfs_nrs_seq_write(struct file *file, const char __user *buffer,
 659                              size_t count, loff_t *off)
 660 {
 661         struct seq_file                *m = file->private_data;
 662         struct ptlrpc_service          *svc = m->private;
 663         enum ptlrpc_nrs_queue_type      queue = PTLRPC_NRS_QUEUE_BOTH;
 664         char                           *cmd;
 665         char                           *cmd_copy = NULL;
 666         char                           *policy_name;
 667         char                           *queue_name;
 668         int                             rc = 0;
 669         ENTRY;
 670
 671         if (count >= LPROCFS_NRS_WR_MAX_CMD)
 672                 GOTO(out, rc = -EINVAL);
 673
 674         OBD_ALLOC(cmd, LPROCFS_NRS_WR_MAX_CMD);
 675         if (cmd == NULL)
 676                 GOTO(out, rc = -ENOMEM);
 677         /**
 678          * strsep() modifies its argument, so keep a copy
 679          */
 680         cmd_copy = cmd;
 681
 682         if (copy_from_user(cmd, buffer, count))
 683                 GOTO(out, rc = -EFAULT);
 684
 685         cmd[count] = '\0';
 686
 687         policy_name = strsep(&cmd, " ");
 688
 689         if (strlen(policy_name) > NRS_POL_NAME_MAX - 1)
 690                 GOTO(out, rc = -EINVAL);
 691
 692         /**
 693          * No [reg|hp] token has been specified
 694          */
 695         if (cmd == NULL)
 696                 goto default_queue;
 697
 698         queue_name = strsep(&cmd, " ");
 699         /**
 700          * The second token is either an optional [reg|hp] string,
 701          * or arguments
 702          */
 703         if (strcmp(queue_name, "reg") == 0)
 704                 queue = PTLRPC_NRS_QUEUE_REG;
 705         else if (strcmp(queue_name, "hp") == 0)
 706                 queue = PTLRPC_NRS_QUEUE_HP;
 707         else {
 708                 if (cmd != NULL)
 709                         *(cmd - 1) = ' ';
 710                 cmd = queue_name;
 711         }
 712
 713 default_queue:
 714
 715         if (queue == PTLRPC_NRS_QUEUE_HP && !nrs_svc_has_hp(svc))
 716                 GOTO(out, rc = -ENODEV);
 717         else if (queue == PTLRPC_NRS_QUEUE_BOTH && !nrs_svc_has_hp(svc))
 718                 queue = PTLRPC_NRS_QUEUE_REG;
 719
 720         /**
 721          * Serialize NRS core lprocfs operations with policy registration/
 722          * unregistration.
 723          */
 724         mutex_lock(&nrs_core.nrs_mutex);
 725
 726         rc = ptlrpc_nrs_policy_control(svc, queue, policy_name,
 727                                        PTLRPC_NRS_CTL_START,
 728                                        false, cmd);
 729
 730         mutex_unlock(&nrs_core.nrs_mutex);
 731 out:
 732         if (cmd_copy)
 733                 OBD_FREE(cmd_copy, LPROCFS_NRS_WR_MAX_CMD);
 734
 735         RETURN(rc < 0 ? rc : count);
 736 }
 737 LPROC_SEQ_FOPS(ptlrpc_lprocfs_nrs);
 738
 739 /** @} nrs */
 740
 741 struct ptlrpc_srh_iterator {
 742         int                     srhi_idx;
 743         __u64                   srhi_seq;
 744         struct ptlrpc_request   *srhi_req;
 745 };
 746
 747 static int
 748 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service_part *svcpt,
 749                                     struct ptlrpc_srh_iterator *srhi,
 750                                     __u64 seq)
 751 {
 752         struct list_head        *e;
 753         struct ptlrpc_request   *req;
 754
 755         if (srhi->srhi_req != NULL &&
 756             srhi->srhi_seq > svcpt->scp_hist_seq_culled &&
 757             srhi->srhi_seq <= seq) {
 758                 /* If srhi_req was set previously, hasn't been culled and
 759                  * we're searching for a seq on or after it (i.e. more
 760                  * recent), search from it onwards.
 761                  * Since the service history is LRU (i.e. culled reqs will
 762                  * be near the head), we shouldn't have to do long
 763                  * re-scans */
 764                 LASSERTF(srhi->srhi_seq == srhi->srhi_req->rq_history_seq,
 765                          "%s:%d: seek seq %llu, request seq %llu\n",
 766                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 767                          srhi->srhi_seq, srhi->srhi_req->rq_history_seq);
 768                 LASSERTF(!list_empty(&svcpt->scp_hist_reqs),
 769                          "%s:%d: seek offset %llu, request seq %llu, "
 770                          "last culled %llu\n",
 771                          svcpt->scp_service->srv_name, svcpt->scp_cpt,
 772                          seq, srhi->srhi_seq, svcpt->scp_hist_seq_culled);
 773                 e = &srhi->srhi_req->rq_history_list;
 774         } else {
 775                 /* search from start */
 776                 e = svcpt->scp_hist_reqs.next;
 777         }
 778
 779         while (e != &svcpt->scp_hist_reqs) {
 780                 req = list_entry(e, struct ptlrpc_request, rq_history_list);
 781
 782                 if (req->rq_history_seq >= seq) {
 783                         srhi->srhi_seq = req->rq_history_seq;
 784                         srhi->srhi_req = req;
 785                         return 0;
 786                 }
 787                 e = e->next;
 788         }
 789
 790         return -ENOENT;
 791 }
 792
 793 /*
 794  * ptlrpc history sequence is used as "position" of seq_file, in some case,
 795  * seq_read() will increase "position" to indicate reading the next
 796  * element, however, low bits of history sequence are reserved for CPT id
 797  * (check the details from comments before ptlrpc_req_add_history), which
 798  * means seq_read() might change CPT id of history sequence and never
 799  * finish reading of requests on a CPT. To make it work, we have to shift
 800  * CPT id to high bits and timestamp to low bits, so seq_read() will only
 801  * increase timestamp which can correctly indicate the next position.
 802  */
 803
 804 /* convert seq_file pos to cpt */
 805 #define PTLRPC_REQ_POS2CPT(svc, pos)                    \
 806         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 807          (__u64)(pos) >> (64 - (svc)->srv_cpt_bits))
 808
 809 /* make up seq_file pos from cpt */
 810 #define PTLRPC_REQ_CPT2POS(svc, cpt)                    \
 811         ((svc)->srv_cpt_bits == 0 ? 0 :                 \
 812          (cpt) << (64 - (svc)->srv_cpt_bits))
 813
 814 /* convert sequence to position */
 815 #define PTLRPC_REQ_SEQ2POS(svc, seq)                    \
 816         ((svc)->srv_cpt_bits == 0 ? (seq) :             \
 817          ((seq) >> (svc)->srv_cpt_bits) |               \
 818          ((seq) << (64 - (svc)->srv_cpt_bits)))
 819
 820 /* convert position to sequence */
 821 #define PTLRPC_REQ_POS2SEQ(svc, pos)                    \
 822         ((svc)->srv_cpt_bits == 0 ? (pos) :             \
 823          ((__u64)(pos) << (svc)->srv_cpt_bits) |        \
 824          ((__u64)(pos) >> (64 - (svc)->srv_cpt_bits)))
 825
 826 static void *
 827 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
 828 {
 829         struct ptlrpc_service           *svc = s->private;
 830         struct ptlrpc_service_part      *svcpt;
 831         struct ptlrpc_srh_iterator      *srhi;
 832         unsigned int                    cpt;
 833         int                             rc;
 834         int                             i;
 835
 836         if (sizeof(loff_t) != sizeof(__u64)) { /* can't support */
 837                 CWARN("Failed to read request history because size of loff_t "
 838                       "%d can't match size of u64\n", (int)sizeof(loff_t));
 839                 return NULL;
 840         }
 841
 842         OBD_ALLOC(srhi, sizeof(*srhi));
 843         if (srhi == NULL)
 844                 return NULL;
 845
 846         srhi->srhi_seq = 0;
 847         srhi->srhi_req = NULL;
 848
 849         cpt = PTLRPC_REQ_POS2CPT(svc, *pos);
 850
 851         ptlrpc_service_for_each_part(svcpt, i, svc) {
 852                 if (i < cpt) /* skip */
 853                         continue;
 854                 if (i > cpt) /* make up the lowest position for this CPT */
 855                         *pos = PTLRPC_REQ_CPT2POS(svc, i);
 856
 857                 spin_lock(&svcpt->scp_lock);
 858                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi,
 859                                 PTLRPC_REQ_POS2SEQ(svc, *pos));
 860                 spin_unlock(&svcpt->scp_lock);
 861                 if (rc == 0) {
 862                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 863                         srhi->srhi_idx = i;
 864                         return srhi;
 865                 }
 866         }
 867
 868         OBD_FREE(srhi, sizeof(*srhi));
 869         return NULL;
 870 }
 871
 872 static void
 873 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
 874 {
 875         struct ptlrpc_srh_iterator *srhi = iter;
 876
 877         if (srhi != NULL)
 878                 OBD_FREE(srhi, sizeof(*srhi));
 879 }
 880
 881 static void *
 882 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
 883                                     void *iter, loff_t *pos)
 884 {
 885         struct ptlrpc_service           *svc = s->private;
 886         struct ptlrpc_srh_iterator      *srhi = iter;
 887         struct ptlrpc_service_part      *svcpt;
 888         __u64                           seq;
 889         int                             rc;
 890         int                             i;
 891
 892         for (i = srhi->srhi_idx; i < svc->srv_ncpts; i++) {
 893                 svcpt = svc->srv_parts[i];
 894
 895                 if (i > srhi->srhi_idx) { /* reset iterator for a new CPT */
 896                         srhi->srhi_req = NULL;
 897                         seq = srhi->srhi_seq = 0;
 898                 } else { /* the next sequence */
 899                         seq = srhi->srhi_seq + (1 << svc->srv_cpt_bits);
 900                 }
 901
 902                 spin_lock(&svcpt->scp_lock);
 903                 rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, seq);
 904                 spin_unlock(&svcpt->scp_lock);
 905                 if (rc == 0) {
 906                         *pos = PTLRPC_REQ_SEQ2POS(svc, srhi->srhi_seq);
 907                         srhi->srhi_idx = i;
 908                         return srhi;
 909                 }
 910         }
 911
 912         OBD_FREE(srhi, sizeof(*srhi));
 913         return NULL;
 914 }
 915
 916 /* common ost/mdt so_req_printer */
 917 void target_print_req(void *seq_file, struct ptlrpc_request *req)
 918 {
 919         /* Called holding srv_lock with irqs disabled.
 920          * Print specific req contents and a newline.
 921          * CAVEAT EMPTOR: check request message length before printing!!!
 922          * You might have received any old crap so you must be just as
 923          * careful here as the service's request parser!!! */
 924         struct seq_file *sf = seq_file;
 925
 926         switch (req->rq_phase) {
 927         case RQ_PHASE_NEW:
 928                 /* still awaiting a service thread's attention, or rejected
 929                  * because the generic request message didn't unpack */
 930                 seq_printf(sf, "<not swabbed>\n");
 931                 break;
 932         case RQ_PHASE_INTERPRET:
 933                 /* being handled, so basic msg swabbed, and opc is valid
 934                  * but racing with mds_handle() */
 935         case RQ_PHASE_COMPLETE:
 936                 /* been handled by mds_handle() reply state possibly still
 937                  * volatile */
 938                 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
 939                 break;
 940         default:
 941                 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
 942         }
 943 }
 944 EXPORT_SYMBOL(target_print_req);
 945
 946 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
 947 {
 948         struct ptlrpc_service           *svc = s->private;
 949         struct ptlrpc_srh_iterator      *srhi = iter;
 950         struct ptlrpc_service_part      *svcpt;
 951         struct ptlrpc_request           *req;
 952         int                             rc;
 953
 954         LASSERT(srhi->srhi_idx < svc->srv_ncpts);
 955
 956         svcpt = svc->srv_parts[srhi->srhi_idx];
 957
 958         spin_lock(&svcpt->scp_lock);
 959
 960         rc = ptlrpc_lprocfs_svc_req_history_seek(svcpt, srhi, srhi->srhi_seq);
 961
 962         if (rc == 0) {
 963                 struct timespec arrival, sent, arrivaldiff;
 964                 char nidstr[LNET_NIDSTR_SIZE];
 965
 966                 req = srhi->srhi_req;
 967
 968                 libcfs_nid2str_r(req->rq_self, nidstr, sizeof(nidstr));
 969                 arrival.tv_sec = req->rq_arrival_time.tv_sec;
 970                 arrival.tv_nsec =
 971                         req->rq_arrival_time.tv_usec * NSEC_PER_USEC;
 972                 sent.tv_sec = req->rq_sent;
 973                 sent.tv_nsec = 0;
 974                 arrivaldiff = timespec_sub(sent, arrival);
 975
 976                 /* Print common req fields.
 977                  * CAVEAT EMPTOR: we're racing with the service handler
 978                  * here.  The request could contain any old crap, so you
 979                  * must be just as careful as the service's request
 980                  * parser. Currently I only print stuff here I know is OK
 981                  * to look at coz it was set up in request_in_callback()!!! */
 982                 seq_printf(s, "%lld:%s:%s:x%llu:%d:%s:"
 983                            "%ld.%06ld:%ld.%06lds(%+ld.0s) ",
 984                            req->rq_history_seq, nidstr,
 985                            libcfs_id2str(req->rq_peer), req->rq_xid,
 986                            req->rq_reqlen, ptlrpc_rqphase2str(req),
 987                            req->rq_arrival_time.tv_sec,
 988                            req->rq_arrival_time.tv_usec,
 989                            arrivaldiff.tv_sec,
 990                            arrivaldiff.tv_nsec / NSEC_PER_USEC,
 991                            req->rq_sent - req->rq_deadline);
 992                 if (svc->srv_ops.so_req_printer == NULL)
 993                         seq_printf(s, "\n");
 994                 else
 995                         svc->srv_ops.so_req_printer(s, srhi->srhi_req);
 996         }
 997
 998         spin_unlock(&svcpt->scp_lock);
 999         return rc;
1000 }
1001
1002 static int
1003 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
1004 {
1005         static struct seq_operations sops = {
1006                 .start = ptlrpc_lprocfs_svc_req_history_start,
1007                 .stop  = ptlrpc_lprocfs_svc_req_history_stop,
1008                 .next  = ptlrpc_lprocfs_svc_req_history_next,
1009                 .show  = ptlrpc_lprocfs_svc_req_history_show,
1010         };
1011         struct seq_file *seqf;
1012         int             rc;
1013
1014         rc = LPROCFS_ENTRY_CHECK(inode);
1015         if (rc < 0)
1016                 return rc;
1017
1018         rc = seq_open(file, &sops);
1019         if (rc)
1020                 return rc;
1021
1022         seqf = file->private_data;
1023         seqf->private = PDE_DATA(inode);
1024         return 0;
1025 }
1026
1027 /* See also lprocfs_rd_timeouts */
1028 static int ptlrpc_lprocfs_timeouts_seq_show(struct seq_file *m, void *n)
1029 {
1030         struct ptlrpc_service           *svc = m->private;
1031         struct ptlrpc_service_part      *svcpt;
1032         struct dhms                     ts;
1033         time64_t worstt;
1034         unsigned int                    cur;
1035         unsigned int                    worst;
1036         int                             i;
1037
1038         if (AT_OFF) {
1039                 seq_printf(m, "adaptive timeouts off, using obd_timeout %u\n",
1040                            obd_timeout);
1041                 return 0;
1042         }
1043
1044         ptlrpc_service_for_each_part(svcpt, i, svc) {
1045                 cur     = at_get(&svcpt->scp_at_estimate);
1046                 worst   = svcpt->scp_at_estimate.at_worst_ever;
1047                 worstt  = svcpt->scp_at_estimate.at_worst_time;
1048                 s2dhms(&ts, ktime_get_real_seconds() - worstt);
1049
1050                 seq_printf(m, "%10s : cur %3u  worst %3u (at %lld, "
1051                            DHMS_FMT" ago) ", "service",
1052                            cur, worst, (s64)worstt, DHMS_VARS(&ts));
1053
1054                 lprocfs_at_hist_helper(m, &svcpt->scp_at_estimate);
1055         }
1056
1057         return 0;
1058 }
1059 LPROC_SEQ_FOPS_RO(ptlrpc_lprocfs_timeouts);
1060
1061 static int ptlrpc_lprocfs_hp_ratio_seq_show(struct seq_file *m, void *v)
1062 {
1063         struct ptlrpc_service *svc = m->private;
1064         seq_printf(m, "%d\n", svc->srv_hpreq_ratio);
1065         return 0;
1066 }
1067
1068 static ssize_t
1069 ptlrpc_lprocfs_hp_ratio_seq_write(struct file *file, const char __user *buffer,
1070                                   size_t count, loff_t *off)
1071 {
1072         struct seq_file *m = file->private_data;
1073         struct ptlrpc_service *svc = m->private;
1074         int rc;
1075         __s64 val;
1076
1077         rc = lprocfs_str_to_s64(buffer, count, &val);
1078         if (rc < 0)
1079                 return rc;
1080
1081         if (val < 0 || val > INT_MAX)
1082                 return -ERANGE;
1083
1084         spin_lock(&svc->srv_lock);
1085         svc->srv_hpreq_ratio = val;
1086         spin_unlock(&svc->srv_lock);
1087
1088         return count;
1089 }
1090 LPROC_SEQ_FOPS(ptlrpc_lprocfs_hp_ratio);
1091
1092 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
1093                                      struct ptlrpc_service *svc)
1094 {
1095         struct lprocfs_vars lproc_vars[] = {
1096                 { .name = "high_priority_ratio",
1097                   .fops = &ptlrpc_lprocfs_hp_ratio_fops,
1098                   .data = svc },
1099                 { .name = "req_buffer_history_len",
1100                   .fops = &ptlrpc_lprocfs_req_history_len_fops,
1101                   .data = svc },
1102                 { .name = "req_buffer_history_max",
1103                   .fops = &ptlrpc_lprocfs_req_history_max_fops,
1104                   .data = svc },
1105                 { .name = "threads_min",
1106                   .fops = &ptlrpc_lprocfs_threads_min_fops,
1107                   .data = svc },
1108                 { .name = "threads_max",
1109                   .fops = &ptlrpc_lprocfs_threads_max_fops,
1110                   .data = svc },
1111                 { .name = "threads_started",
1112                   .fops = &ptlrpc_lprocfs_threads_started_fops,
1113                   .data = svc },
1114                 { .name = "timeouts",
1115                   .fops = &ptlrpc_lprocfs_timeouts_fops,
1116                   .data = svc },
1117                 { .name = "nrs_policies",
1118                   .fops = &ptlrpc_lprocfs_nrs_fops,
1119                   .data = svc },
1120                 { NULL }
1121         };
1122         static struct file_operations req_history_fops = {
1123                 .owner       = THIS_MODULE,
1124                 .open        = ptlrpc_lprocfs_svc_req_history_open,
1125                 .read        = seq_read,
1126                 .llseek      = seq_lseek,
1127                 .release     = lprocfs_seq_release,
1128         };
1129
1130         int rc;
1131
1132         ptlrpc_lprocfs_register(entry, svc->srv_name,
1133                                 "stats", &svc->srv_procroot,
1134                                 &svc->srv_stats);
1135         if (svc->srv_procroot == NULL)
1136                 return;
1137
1138         lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
1139
1140         rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
1141                                 0400, &req_history_fops, svc);
1142         if (rc)
1143                 CWARN("Error adding the req_history file\n");
1144 }
1145
1146 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
1147 {
1148         ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
1149                                 &obddev->obd_svc_procroot,
1150                                 &obddev->obd_svc_stats);
1151 }
1152 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
1153
1154 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req, long amount)
1155 {
1156         struct lprocfs_stats *svc_stats;
1157         __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
1158         int opc = opcode_offset(op);
1159
1160         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1161         if (svc_stats == NULL || opc <= 0)
1162                 return;
1163         LASSERT(opc < LUSTRE_MAX_OPCODES);
1164         if (!(op == LDLM_ENQUEUE || op == MDS_REINT))
1165                 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, amount);
1166 }
1167
1168 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int bytes)
1169 {
1170         struct lprocfs_stats *svc_stats;
1171         int idx;
1172
1173         if (!req->rq_import)
1174                 return;
1175         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
1176         if (!svc_stats)
1177                 return;
1178         idx = lustre_msg_get_opc(req->rq_reqmsg);
1179         switch (idx) {
1180         case OST_READ:
1181                 idx = BRW_READ_BYTES + PTLRPC_LAST_CNTR;
1182                 break;
1183         case OST_WRITE:
1184                 idx = BRW_WRITE_BYTES + PTLRPC_LAST_CNTR;
1185                 break;
1186         default:
1187                 LASSERTF(0, "unsupported opcode %u\n", idx);
1188                 break;
1189         }
1190
1191         lprocfs_counter_add(svc_stats, idx, bytes);
1192 }
1193
1194 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
1195
1196 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
1197 {
1198         if (svc->srv_procroot != NULL)
1199                 lprocfs_remove(&svc->srv_procroot);
1200
1201         if (svc->srv_stats)
1202                 lprocfs_free_stats(&svc->srv_stats);
1203 }
1204
1205 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
1206 {
1207         if (obd->obd_svc_procroot)
1208                 lprocfs_remove(&obd->obd_svc_procroot);
1209
1210         if (obd->obd_svc_stats)
1211                 lprocfs_free_stats(&obd->obd_svc_stats);
1212 }
1213 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
1214
1215 ssize_t
1216 lprocfs_ping_seq_write(struct file *file, const char __user *buffer,
1217                        size_t count, loff_t *off)
1218 {
1219         struct seq_file         *m = file->private_data;
1220         struct obd_device       *obd = m->private;
1221         struct ptlrpc_request   *req;
1222         int                     rc;
1223         ENTRY;
1224
1225         LPROCFS_CLIMP_CHECK(obd);
1226         req = ptlrpc_prep_ping(obd->u.cli.cl_import);
1227         LPROCFS_CLIMP_EXIT(obd);
1228         if (req == NULL)
1229                 RETURN(-ENOMEM);
1230
1231         req->rq_send_state = LUSTRE_IMP_FULL;
1232
1233         rc = ptlrpc_queue_wait(req);
1234
1235         ptlrpc_req_finished(req);
1236         if (rc >= 0)
1237                 RETURN(count);
1238         RETURN(rc);
1239 }
1240 EXPORT_SYMBOL(lprocfs_ping_seq_write);
1241
1242 /* Write the connection UUID to this file to attempt to connect to that node.
1243  * The connection UUID is a node's primary NID. For example,
1244  * "echo connection=192.168.0.1@tcp0::instance > .../import".
1245  */
1246 ssize_t
1247 lprocfs_import_seq_write(struct file *file, const char __user *buffer,
1248                          size_t count, loff_t *off)
1249 {
1250         struct seq_file   *m    = file->private_data;
1251         struct obd_device *obd  = m->private;
1252         struct obd_import *imp  = obd->u.cli.cl_import;
1253         char *kbuf = NULL;
1254         char *uuid;
1255         char *ptr;
1256         int do_reconn = 1;
1257         const char prefix[] = "connection=";
1258         const int prefix_len = sizeof(prefix) - 1;
1259
1260         if (count > PAGE_SIZE - 1 || count <= prefix_len)
1261                 return -EINVAL;
1262
1263         OBD_ALLOC(kbuf, count + 1);
1264         if (kbuf == NULL)
1265                 return -ENOMEM;
1266
1267         if (copy_from_user(kbuf, buffer, count))
1268                 GOTO(out, count = -EFAULT);
1269
1270         kbuf[count] = 0;
1271
1272         /* only support connection=uuid::instance now */
1273         if (strncmp(prefix, kbuf, prefix_len) != 0)
1274                 GOTO(out, count = -EINVAL);
1275
1276         uuid = kbuf + prefix_len;
1277         ptr = strstr(uuid, "::");
1278         if (ptr) {
1279                 __u32 inst;
1280                 char *endptr;
1281
1282                 *ptr = 0;
1283                 do_reconn = 0;
1284                 ptr += 2; /* Skip :: */
1285                 inst = simple_strtol(ptr, &endptr, 10);
1286                 if (*endptr) {
1287                         CERROR("config: wrong instance # %s\n", ptr);
1288                 } else if (inst != imp->imp_connect_data.ocd_instance) {
1289                         CDEBUG(D_INFO, "IR: %s is connecting to an obsoleted "
1290                                "target(%u/%u), reconnecting...\n",
1291                                imp->imp_obd->obd_name,
1292                                imp->imp_connect_data.ocd_instance, inst);
1293                         do_reconn = 1;
1294                 } else {
1295                         CDEBUG(D_INFO, "IR: %s has already been connecting to "
1296                                "new target(%u)\n",
1297                                imp->imp_obd->obd_name, inst);
1298                 }
1299         }
1300
1301         if (do_reconn)
1302                 ptlrpc_recover_import(imp, uuid, 1);
1303
1304 out:
1305         OBD_FREE(kbuf, count + 1);
1306         return count;
1307 }
1308 EXPORT_SYMBOL(lprocfs_import_seq_write);
1309
1310 int lprocfs_pinger_recov_seq_show(struct seq_file *m, void *n)
1311 {
1312         struct obd_device *obd = m->private;
1313         struct obd_import *imp = obd->u.cli.cl_import;
1314
1315         LPROCFS_CLIMP_CHECK(obd);
1316         seq_printf(m, "%d\n", !imp->imp_no_pinger_recover);
1317         LPROCFS_CLIMP_EXIT(obd);
1318         return 0;
1319 }
1320 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_show);
1321
1322 ssize_t
1323 lprocfs_pinger_recov_seq_write(struct file *file, const char __user *buffer,
1324                                size_t count, loff_t *off)
1325 {
1326         struct seq_file *m = file->private_data;
1327         struct obd_device *obd = m->private;
1328         struct client_obd *cli = &obd->u.cli;
1329         struct obd_import *imp = cli->cl_import;
1330         int rc;
1331         __s64 val;
1332
1333         rc = lprocfs_str_to_s64(buffer, count, &val);
1334         if (rc < 0)
1335                 return rc;
1336
1337         if (val != 0 && val != 1)
1338                 return -ERANGE;
1339
1340         LPROCFS_CLIMP_CHECK(obd);
1341         spin_lock(&imp->imp_lock);
1342         imp->imp_no_pinger_recover = !val;
1343         spin_unlock(&imp->imp_lock);
1344         LPROCFS_CLIMP_EXIT(obd);
1345         return count;
1346 }
1347 EXPORT_SYMBOL(lprocfs_pinger_recov_seq_write);
1348
1349 #endif /* CONFIG_PROC_FS */